Cuckoo: Towards Decentralized, Socio-Aware Online Microblogging Services and Data Measurements
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Cuckoo: Towards Decentralized, Socio-Aware Online Microblogging Services and Data Measurements Tianyin Xu1,2 , Yang Chen1 , Jin Zhao1,3 , Xiaoming Fu1 1 Institute of Computer Science, University of Goettingen, Goettingen, Germany 2 State Key Lab. for Novel Software and Technology, Nanjing University, Nanjing, China 3 School of Computer Science, Fudan University, Shanghai, China {tianyin.xu, yang.chen, fu}@cs.uni-goettingen.de, jzhao@fudan.edu.cn ABSTRACT 1. INTRODUCTION Online microblogging services, as exemplified by Twitter [9] With the phenomenal success of Twitter-like Internet ser- and Yammer [12], have become immensely popular during vices, microblogging (e.g., Twitter [9], Yammer [12], Buzz the latest three years. Twitter, the most successful mi- [5], etc) has emerged as a new, popular communication util- croblogging service, has attracted more than 41.7 million ity in the latest years. Microblogging has become popular users as of July 2009 [25] and is still growing fast. However, quite quickly. For example, Twitter, the most successful mi- current microblogging systems severely suffer from perfor- croblogging service launched in October 2006, has attracted mance bottlenecks and central points of failure due to their more than 41.7 million users as of July 2009 and its userbase centralized architecture. Thus, centralized microblogging is still growing incredibly fast. Nowadays, microblogging systems may threaten the scalability, reliability, as well as services play important roles in many aspects of people’s availability of the offered services, not to mention the ex- social life [21, 35]: from releasing emotional stress, to keep- tremely high operational and maintenance cost. ing in touch with friends, to collaborating with colleagues However, it is not trivial to decentralize microblogging ser- at work, etc. Moreover, the functionalities of microblogging vices in a peer-to-peer fashion. The challenges first derive services are even beyond human social networks. Microblog- from the heterogeneity of the inherent online social network ging is becoming an important media for spreading news to (OSN) features. The non-reciprocation feature of microblog- large groups of people [25, 31, 35], from new product mar- ging services also increases the heterogeneity. Moreover, dif- keting, to political campaign publicity, to immediate news ferent from traditional approaches used in centralized server- breaking, etc. One impressive story is that Twitter broke based systems, an efficient, robust and scalable approach for the news about China’s massive earthquake well before the data collection and dissemination in such distributed hetero- mainstream media [11]. geneous environments is desirable. Current microblogging services depend on centralized ar- In this paper, we present a decentralized, socio-aware mi- chitectures, where user clients repeatedly request centralized croblogging system named Cuckoo. The design takes advan- servers (e.g., Twitter Server) for newest micro-contents no tages of the inherent social relationships while leverages P2P matter whether there is a new update1 . Such polling-style techniques towards scalable, reliable microblogging services. requests are blind, superfluous and sticky. The centralized Besides, Cuckoo provides a flexible interface for data collec- microblogging systems suffer severely from performance bot- tion while circumventing unnecessary traffic on the server. tleneck (e.g., Twitter’s rate limiting, 150 requests per hour is We discuss the benefits that our system may bring for both allowed) and central point of failure (e.g., Twitter’s not rare service providers and end users. We also discuss the techni- “over capacity” errors and “database maintenance” errors). cal aspects to be considered and report our work in progress. Although Twitter has made several performance improve- ments, such as removing the upper limit on the subscrip- tion number in 2009, by increasing the number of dedicated Categories and Subject Descriptors servers, it still can hardly catch up with the steadily-growing C.2.4 [Computer-Communication Networks]: Distributed user demands. As a result, centralized microblogging sys- Systems—Distributed Applications tems may threaten the scalability, reliability, as well as avail- ability of the offered services, not to mention the extremely General Terms high operational and maintenance cost. Design, Performance One important aspect is data collection. Data collection is critical for researchers and third-party developers to un- Keywords derstand user behavior and access patterns of microblogging Microblogging Services, Online Social Networking services. So far, data collection is performed via data crawl- ing on centralized servers [21, 25], which may cause traffic storms and is not scalable. Even worse, such methods may Permission to make digital or hard copies of all or part of this work for fail to retrieve statistics from departed peers. More efficient personal or classroom use is granted without fee provided that copies are and scalable data collection schemes are required to let users not made or distributed for profit or commercial advantage and that copies directly exchanging vital statistics in a decentralized fashion. bear this notice and the full citation on the first page. To copy otherwise, to However, it is not trivial to decentralize microblogging ser- republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. 1 HotPlanet’10 , June 15, 2010, San Francisco, CA, USA In this paper, we use “microblog”, “micro-content”, “micro- Copyright 2010 ACM 978-1-4503-0177-0 ...$10.00. news”, “status”, “update” interchangeably.
vices in a peer-to-peer fashion. The challenges mainly derive 2. BACKGROUND AND RELATED WORK from the heterogeneity of the inherent online social network Thanks to the success of Twitter [9], a number of mi- (OSN) features. Furthermore, the non-reciprocation feature croblogging services are developed and deployed on the In- of microblogging’s opt-in subscription model [31] increases ternet [5,6,8,12]. Although each of them has some additional the heterogeneity. The basic operation of the subscription features than original Twitter, common to these microblog- model is “follow”. Being a follower means that the user will ging services is the opt-in subscription model based on the receive all the micro-contents from the one he follows. A “follow” operation. The subscription model is deceptively user can follow anyone, and the one being followed need not simple. A user can publish microblogs with a length limit of follow back. We summarize the main challenges as follows: viewable text (up to 140 characters in Twitter). The other users who have explicitly follow that user will receive this 1. Support users with heterogenous social relationships, microblog, i.e., being a follower means that the user will such as broadcasters and miscreants [23]. Broadcast- receive all the micro-contents from the one he follows. A ers include celebrities and news media that have huge user can follow anyone, and the one being followed need not numbers of followers. For example, cnnbrk (CNN follow back. Note that the relationship of following and be- Breaking News) in Twitter has over 2.97 million follow- ing followed requires no reciprocation, i.e., “follow” is a uni- ers but only follows 28 users. Miscreants try to contact directional operation. This opt-in, social networking model everyone and hope that someone can follow back. The effectively resists spamming so that users only receive the design should be able to support different user social microblogs from who they are concerning. On the other patterns with different purpose of use. hand, the “small-world” social feature [28] makes it possible 2. Support users with heterogenous access patterns as for a free comment to blossom into a worldwide discussion. well as social behaviors. Unlike other P2P applica- The popularity of microblogging services has already at- tion such as file downloading and video streaming, mi- tracted some research interests in the computer networking croblogging users have highly dynamic access patterns. perspective [21,23,25,31]. Most of these works are measure- In case of Twitter, various communication channels ment studies on Twitter including Twitter’s social graph, are supported including SMS, RSS feeds, Web, dedi- its topological and geographical characteristics, the popu- cated software, etc. Moreover, the users’ behaviors are larity of the micro-contents, etc. Java et al. [21] report some quite different. Such heterogeneity causes significant early observations of Twitter in 2007. They study the topo- dynamics to decentralized systems (e.g., peer churn). logical and geographical properties and show user clusters formed by users with similar intention. Krishnamurthy et 3. Support users with efficient, robust and scalable data al. [23] identify distinct classes of Twitter users as well as collection. Since the social profiles/attributes and other their behaviors according to the number of followings and interaction information are very important aspects for followers. Kwak et al. [25] crawl the entire Twittersphere system designers and socialists to understand the user based on which a comprehensive measurement study is per- behaviors, restore from failures or peer churns as well formed. In the following-follower analysis, they find a no- as improve the system design, it is crucial to collect power-law follower distribution, a short effective diameter, related information in an efficient, robust and scalable and low reciprocity. The influentials, trending topics, and manner, whilst protecting the users’ privacy. impact of retweet are also studied. The only proposed decentralized system prototype focus- In this position paper, we present a decentralized, socio- ing on microblogging so far is FETHR [31], which is most aware online microblogging system named Cuckoo. Our ob- related to our work. FETHR [31] envisions the fully de- jective is to design, implement, and deploy a decentralized centralized microblogging for the first time. The main idea socio-aware microblogging service, which is fully compatible of FETHR is to let users directly contact each other via to the current Twitter architecture. In other words, we ex- HTTP and employ gossip for popular micro-content propa- pect to study how the decentralized Cuckoo system helps gation. However, FETHR lacks of several practical consider- microblogging service to improve the performance, to save ations, e.g., decentralized user lookup, overlay maintenance, the bandwidth costs, and to reduce the sluggishness and etc. In addition, FETHR cannot provide any guarantee on downtime, while performing equally well or even better than micro-content delivery and the delivery performance will be the conventional centralized system. For this objective, the degraded under asynchronous user access patterns. design of Cuckoo conserves valuable bandwidth and storage There are some decentralized OSN prototypes which are resources on Twitter server while taking advantage of peer related to our work [15, 17, 24, 32]. PeerSoN [15] is a project assistance in a complementary fashion. In response to the for P2P OSNs that proposes to use dedicated third-party challenges, Cuckoo takes advantages of social relationship DHTs for lookup meta-data. Based on the meta-data, di- while leverages P2P techniques towards scalable, reliable mi- rect information exchange between users can be achieved. croblogging services. Different from traditional approaches Safebook [17] aims at protecting users’ security and privacy. used in centralized server-based systems [21,25], data col- In Safebook, the “matryoshka” structures are designed to lection and dissemination in such distributed heterogeneous protect users data based on trust transitivity. Another re- environments are challenged by the dynamics, heterogeneity lated decentralized OSN system, Vis-à-Vis [32] is based on and distributed nature of the system. the concept of Virtual Individual Server (VIS), which is used The rest of the paper is organized as follows. Sec. 2 re- for managing and storing user data. VISs self-organized into views the background and related works. Sec. 3 presents the multi-tier DHTs that represent OSN groups, with one DHT design rationale of Cuckoo and summarizes the incentives per group. The top-tier meta-DHT is used to advertise and of Cuckoo adoption which may brings for both the service search for public OSN groups. uaOSN [24] proposes to dis- provider and the end users. We discuss the technical aspects tribute only the storage of heavy content while retain the to be considered and report our work in progress in Sec. 4. business model and functionalities of OSN provides. Sec. 5 concludes the paper and poses the future work.
Table 1: Comparison with the related systems System Design Target Socio-Aware Compatible Additional Infr. Struc. Overlay √ √ √ Cuckoo microblogging × FETHR microblogging × × × √ × √ PeerSoN OSN × √ × √ Safebook OSN × × √ √ Vis-à-Vis OSN × × √ √ uaOSN OSN × × These works give good inspiration to the design of Cuckoo. reported in [31] that half of Twitter users have 10 or fewer However, since previous works do not consider the features of followers), he keeps connection with all his followers. Oth- microblogging service, they cannot fit the new service well. erwise for a user has huge numbers of followers, he keeps We summarize these works in comparison with Cuckoo in connections with a logarithmic subset of all the followers. Table 1. The system features listed for comparison include Finally, the neighbor list is initialized during the bootstrap- the target service of design (Design Target), whether tak- ping stage. In boostrapping, the user updates its following’s ing advantage of social relationship (Socio-Aware), whether recent statuses and initializes the neighbor list in this pro- compatible for current system (Compatible), whether re- cess. quiring additional infrastructure (e.g., third-party DHT [15]), 001ab0 and whether based on structured overlay. Virtual Node d46a1c friend friend 3. DESIGN RATIONALE d462ba 420a1b In typical microblogging services, a user has one or several DHT route friend of the following social relations: friend, neighbor, follower, follower DHT and following. Friend is a reciprocate social link between two route neighbor users, which indicates that two users are acquaint with each d4213f 64fb26 other and willing to help each other. Microblogging (e.g., DHT 65a1fc Twitter) shows a quite low level of such reciprocity com- route DHT route pared with conventional OSNs [25]. Neighbor refers to the relationship between users with common interests. For ex- ample, two users sharing a same following are neighbors. In d13da3 microblogging, follower and following are the most common one-way connections. Our design takes advantage of these social relationship to construct the decentralized service. For providing location service and improving availability, Figure 1: Network Model Cuckoo organizes user clients into a structured P2P over- lay. Without loss of generality, we utilize Pastry [30] as the underlying P2P overlay. The 128-bit nodeId can be generated according to the sequential userId (e.g., Twitter 3.2 Update Recent Statuses userId). In this case, a user can find any online user in less When joining the system, a new node setups its Pastry than ⌈log2b N ⌉ steps on average (N is the number of online routing table and then utilizes the underlying infrastructure nodes and b is a configuration parameter with typical value to fetch the updated statuses of its followings. Note that if 4). In our current design, we do not let peers store irrelevant the following user is popular2 , it is likely that the routing micro-contents for security and fairness considerations. path contains the new node’s neighbors. In this case, the routing process will not continue. The new neighbor directly 3.1 Node State sends the recent statuses of the following to the new node Besides the Pastry routing table, each user maintains 4 while help to initialize the neighbor list. Fig. 1 gives an lists for friends, neighbors, followings, and followers. A example for this process, in which node 65a1fc gets the user keeps connection with m online friends. We set m = statuses of node d46a1c directly from node d4213f. Even max(min(f, T ), log2b f ), where f is the user’s number of if no neighbor is met, the request message will reach the friends and T is a threshold. The user itself and his f friends destination and both recent statuses and neighbor list can make up the virtual node (VN) via request redirection, i.e., be updated. Assume the popularity of the following user friends help each other to balance load and improve avail- is δ, the expected ∑ steps for i−1 fetching the updated statuses is ability. Fig. 1 gives an example of the VN mechanism. The approximately K i=1 (1 − δ) · δ · i (K = ⌈log2b N ⌉). VN d46a1c consists of the physical node d46a1c and its Note that DHT is efficient for locating rare items but in- 3 friends (001ab0, 420a1b, 64fb26). The following list curs higher overheads. On the other side, flooding-based is a static config file containing nodeId of all the follow- searching is more efficient for locating highly popular items ing users. No real-time connection is required. Like the but poorly suited for locating rare items [27, 34]. Thus, our friend list, a user also maintains n random followers where design chooses a hybrid searching methods, i.e., use flood- n = max(min(l, T ′ ), log2b l), (l is the number of followers ing to fetch statuses from influentials like cnnbrk and use and T ′ is a threshold). The idea of the calculation of m 2 Here popularity refers to the ratio between the number of and n is that if a user has 10 or fewer followers/friends (it is followers and the total number of users.
DHT to fetch statuses from normal users. The flooding is through “near” nodes according to some proximity metric. Fortunately, Pastry have already provided these near nodes Service Provider SMS in its “neighborhood set”. SMS 3.3 Micro-content Propagation SMS When an online user publish a new micro-content, the microblogging service should disseminate it to all the user’s Mobile Users followers. We use push method instead of conventional pull upload method to achieve the dissemination. The push method is much more efficient than the pull method. It does not require meta-data exchanging or periodical requests. More- DHT over, by replicating micro-contents among followers, the fol- gossip gossip Ring lowers can help each other in a cache-and-relay style, in the gossip case the original publisher later become unavailable or the publisher cannot afford sending updates to all his followers. Internet Users For normal users, directly pushing messages from an orig- inal publisher to his followers seems to be enough (it is re- ported that 90% users in Twitter have less than 100 follow- Figure 2: System Architecture ers [31]). For the broadcasters or influentials like cnnbrk, the publisher cannot afford sending updates to all his follow- ers. In this case, our design uses gossip-based push between Cuckoo in today’s Internet environment, although our main neighbors to propagate the micro-news. Gossip (also called focus is not on security. The presence of malicious users and epidemic, rumor) protocol has been proved to be a robust malwares in the overlay network requires additional security and scalable way of propagating information in dynamic and mechanism to safeguard against attack. heterogeneous networks [19]. The theoretical support pro- One important thing is authentication since some malware vided in [22] proves if there are n nodes and each node gos- may impersonate some normal users to distribute spam. To sips to log(n)+k other nodes on average, the probability that defend against the forged mico-contents, asymmetric key −k everyone gets the message converges to e−e , very close to cryptography and digital signature can be employed, i.e., 1.0 without considering the bandwidth constraint, latency, publishers include digital signatures with each microblogs failure, etc. This result provides a guideline for the design of or even encrypt the micro-contents. To defend against the membership management, i.e., maintain the neighbor list to forged users, some sophisticated authentication infrastruc- be logarithmic in the size of the number of all the followers. ture should be used, e.g., challenge-response protocol. Another problem is suppression, i.e., malicious interme- 3.4 Functionalities of Service Providers diate nodes refuse to forward micro-contents or filter some In our system, if an unpopular node is not online nor its important micro-news. FETHR [31] proposes to link a user’s friends, its profile and recent statuses cannot be found in the microblogs together into a hash chain according to the time- overlay network. In this case, the dedicated servers from line, based on which the integrity can be checked. the service provider (e.g., Twitter server) work as backup The third problem is privacy preserving. Although cur- servers. The omnipotent servers are always online and guar- rent microblogging services like Twitter and Identi.ca seem antee high availability of the service. Note that the users quite open, there are still some users who do not want their fetch profiles and statuses from servers only when they can- profiles and microblogs to be in public. In this case, data en- not find them successfully in the overlay network. In most cryption and control access by appropriate key sharing and cases, it is mainly due to the inactive and unpopular users, distribution is necessary There are already some great effort and thus requesting from servers causes acceptable overhead. to protect privacy in OSNs including the PKI-based (public When publishing new microblogs, the publisher directly key infrastructure) approach [15] and attribute-based en- uploads the digest of the microblog to the server in the same cryption (ABE) [14]. time of gossiping. In our design, we allow long blog con- tent (with more characters than 140) exchanging within the 3.6 Data Collection overlay network. On the other hand, the users only upload One important aspect of microblogging systems is data fixed-length (e.g., 140 characters) digests to the server, with collection which is critical for researchers and third-party the consideration of the micro features that is suitable for developers to understand user behaviors as well as user ac- SMS transmission. cess patterns. So far, data collection is performed via crawl- Fig. 2 demonstrates the system architecture as well as the ing on centralized servers [21, 25], which may cause traffic relationship between service provider and end users. Our de- storms and is not scalable. Even worse, such methods fail sign does not exclude the service provider from the picture. to retrieve statistics from departed peers. In Cuckoo, data Instead of decentralizing the full system, we propose to en- collectors are able to collect vital statistics based on the rich the users’ quality of experience in the overlay network underlying overlay network so as to circumvent server bot- while let the servers provide high availability and reliability. tleneck. The basic idea is to let peers directly exchange In other words, our objective is to help the service providers statistics in a decentralized fashion. Advanced techniques (i.e., the business companies) but no to bury them. such as network coding [29] and data aggregation [33] can be integrated. 3.5 Security Issues It is worth noticing that in microblogging services as well We would be remiss if not discuss the security issues of as OSNs, the developers should collect and measure social
information to improve system design, in addition to the tra- key-based routing (KBR) API defined in [18]. Vis-à-Vis [32] ditional network metrics (e.g., end-to-end delay [13], avail- and uaOSN [24] propose to use some more stable external able bandwidth [20], transmission error [16]). The social storage facilities (Amazon EC2 [1] and Amazon S3 [2] in Vis- information includes user popularity, user social behavior, à-Vis, set-top boxes/residential routers with hard drives [26] user access pattern, etc. Cuckoo includes an inherent data in uaOSN) to achieve high availability. However, the former collection module which allows the efficient, robust and scal- is too expensive for end users (running a virtual machine able exchange, look up and storage of such social data in a for one month costs close to US$75), while the latter is re- privacy protected manner. Further details are currently in- stricted by the deployment difficulties. Our implementation vestigated and extended for potentially other OSNs. just employs end users’ desktop machines to construct the decentralized system. We are confident to provide good per- 3.7 Incentives formance in the face of peer churn. On the other hand, we We summarize the incentives of adoption of Cuckoo in two can expect a much higher performance when the expensive sides: the service provider side and the end user side. For external storage facilities are available. the service provider, the incentives of adopting the Cuckoo In the next stage, we would like to validate our design system is straightforward: and study the performance gain by deploying the system on PlanetLab [7] and testing it by real microblogging traces 1. Lower Bandwidth Cost. Servers can get rid of su- and dataset, e.g., the dataset provided in [25]. We also plan perfluous, sticky and repeated HTTP requests from to implement a Twitter client software that is based on the huge numbers of user clients even there is no new up- design of Cuckoo and is fully compatible to current Twitter’s date. Bandwidth is mainly used for collecting users’ microblogging service. Note that Cuckoo does not require profiles and microblogs. any functionality and modification on the server side. We expect to collect large volumes of microblogging user data 2. High Scalability. The decentralized architecture is via this Cuckoo-based Twitter client. highly scalable that moderates service provider’s grow- ing pain in the face of steadily growing userbase. Less infrastructures are demanded compared with the cen- 5. CONCLUSIONS AND FUTURE WORK tralized equivalents. Although in the infancy, microblogging services are having the golden era in the latest years. Twitter, the pioneer mi- 3. Higher Security. The decentralized nature makes croblogging service provider has attracted more than 41.7 it more robust to malicious attacks, e.g., DoS attacks million users in less than three years and its userbase is (Twitter did be a victim of DDoS attack [10]). still growing incredibly fast. However, current microblog- ging systems severely suffer from performance bottlenecks 4. Better Quality of Service. Better QoS can be and central points of failure due to the unscalable central- achieved by removing the performance bottleneck and ized architecture. In this position paper, we present a decen- single point of failure. tralized, socio-aware microblogging system, named Cuckoo. We present the design rationale of Cuckoo that takes advan- Most important, the business company will not lose any tages of the inherent social relationships while leverages P2P functionality nor the user community. For the end users, the techniques towards scalable, reliable microblogging services. incentives include: Besides, Cuckoo provides a flexible interface for data collec- tion while circumventing unnecessary traffic on the server. 1. High Reliability. Since both user profiles and mi- We discuss the technical aspects to be considered and show croblogs are distributed and replicated in the distributed that Cuckoo can bring good incentives of adoption for both overlay network in addition to the central server, infor- service providers and end users. mation losses in a single location can be easily restored. In the future, we plan to improve the work in the following Besides, the failure of central server (e.g., outrage [3]) aspects. First, we expect to support “topic trend” function will not paralyze the whole system. in Cuckoo like the “#” operation in Twitter. A quite com- 2. Better Quality of Experience. Corresponding to mon use for microblogging services these days is looking at better QoS offered by the service provider, end users particular topics (e.g., UK general election). Second, we can receive better QoE including quick response time, are interested in supporting user mobility in Cuckoo. Cur- higher searching efficiency, longer contents allowed, etc. rently, Cuckoo is designed only for desktop Internet users (the SMS is provided by dedicate servers). User mobility 3. Enrichment of Additional Functions. Cuckoo’s via wireless Internet access implies high churn and should decentralized overlay network is an open, powerful in- be designed carefully. Third, we hope to enrich more third- frastructure for third-party developers to enrich addi- party functions in Cuckoo such as group communication, tional functions for end users which are not yet pro- real-time chatting in addition to the basic “follow” opera- vided by service providers, e.g., photo and multimedia tion. Fourth, we would like to drive the service providers contents in microblogs. to add some functions on the server side in a complemen- tary manner that can benefit the whole system. Finally, 4. IMPLEMENTATION the details about data collection module of Cuckoo will be We are now working on to implement the proposed mi- explored, which is expected to be applicable for other OSNs. croblogging system prototype. We choose FreePastry [4] as our overlay infrastructure for Pastry’s good properties 6. ACKNOWLEDGEMENT (e.g., locality aware) as well as FreePastry’s platform inde- The authors would like to thank Pan Hui from Deutsche pendence (Java source code). Cuckoo can be directly ap- Telekom Laboratories and Tristan Henderson from Univer- plied by any structured overlay network that supports the sity of St Andrews for their helpful comments.
7. REFERENCES Probabilistic Reliable Dissemination in Large-Scale [1] Amazon Elastic Compute Cloud (EC2). Systems. IEEE Transactions on Parallel and http://aws.amazon.com/ec2/. Distributed Systems, 14(3):248–258, 2003. [2] Amazon Simple Storage Service (S3). [23] B. Krishnamurthy, P. Gill, and M. Arlitt. A Few http://aws.amazon.com/s3/. Chirps About Twitter. In WOSN’08: Proc. of the 1st [3] Facebook database outrage cut off about 150,000. ACM SIGCOMM Workshop on Online Social http://news.cnet.com/8301-13577_3-10373349- Networks, Seattle, WA, USA, August 2008. 36.html/. [24] M. Kryczka, R. Cuevas, C. Guerrero, E. Yoneki, and [4] FreePastry. http://www.freepastry.org/. A. Azcorra. A First Step Towards User Assisted [5] Google Buzz. http://www.google.com/buzz/. Online Social Networks. In SNS’10: Proc. of the 2nd [6] Identi.ca – public timeline. http://identi.ca/. ACM Workshop on Social Network Systems, Paris, [7] PlanetLab. http://www.planetlab.org/. France, April 2010. [8] Plurk is a social journal for your life. [25] H. Kwak, C. Lee, H. Park, and S. Moon. What is http://www.plurk.com/. Twitter, a Social Network or a News Media? In [9] Twitter. http://www.twitter.com/. WWW’10: Proc. of the 19th International World [10] Twitter, Facebook attack targeted one user. Wide Web Conference, Raleigh, NC, USA, April 2010. http://news.cnet.com/8301-27080_3-10305200- [26] N. Laoutaris, P. Rodriguez, and L. Massoulié. 245.html?tag=mncol. ECHOS: Edge Capacity Hosting Overlays of Nano [11] ‘Twitters’ beat media in reporting china earthquake. Data Centers. ACM SIGCOMM Computer http://afp.google.com/article/ALeqM5hUZGbUECMKx Communication Review, 38(1):51–54, 2008. _P5H3lRXPnGo6cPjw/. [27] B. T. Loo, R. Huebsch, I. Stoica, and J. M. [12] Yammer: Enterprise Microblogging. Hellerstein. The Case for a Hybrid P2P Search https://www.yammer.com/. Infrastructure. In IPTPS’04: Proc. of the 3rd International Workshop on Peer-to-Peer Systems, San [13] S. Agarwal and J. R. Lorch. Matchmaking for Online Diego, CA, USA, February 2004. Games and Other Latency-Sensitive P2P Systems. In SIGCOMM’09: Proc. of the 15th ACM SIGCOMM, [28] S. Milgram. The Small-World Problem. Psychology Barcelona, Spain, August 2009. Today, 1(1):61–67, 1967. [14] R. Baden, A. Bender, N. Spring, B. Bhattacharjee, [29] D. Niu and B. Li. Circumventing Server Bottlenecks: and D. Starin. Persona: An Online Social Network Indirect Large-Scale P2P Data Collection. In with User-Defined Privacy. In SIGCOMM’09: Proc. of ICDCS’08: Proc. of the 28th International Conference the 15th SIGCOMM, Barcelona, Spain, August 2009. on Distributed Computing Systems, Beijing, China, June 2008. [15] S. Buchegger, D. Schiöberg, L.-H. Vu, and A. Datta. PeerSoN: P2P Social Networking – Early Experiences [30] A. Rowstron and P. Druschel. Pastry: Scalable, and Insights. In SNS’09: Proc. of the 2nd ACM decentralized object location and routing for Workshop on Social Network Systems, Nuremberg, large-scale peer-to-peer systems. In Middleware’01: Germany, March 2009. Proc. of the 18th IFIP/ACM International Conference on Distributed System Platforms, Heidelberg, [16] D. D. Couto, D. Aguayo, J. Bicket, and R. Morris. A Germany, November 2001. High-Throughput Path Metric for Multi-Hop Wireless Routing. In MobiCom’03: Proc. of the 10th ACM [31] D. R. Sandler and D. S. Wallach. Birds of a FETHR: MobiCom, San Diego, CA, September 2003. Open, decentralized micropublishing. In IPTPS’09: Proc. of the 8th International Workshop on [17] L. A. Cutillo, R. Molva, and T. Strufe. Safebook: a Peer-to-Peer Systems, Boston, MA, USA, April 2009. Privacy Preserving Online Social Network Leveraging on Real-Life Trust. IEEE Communication Magazine, [32] A. Shakimov, A. Varshavsky, L. P. Cox, and 47(12):94–101, 2009. R. Cáceres. Privcy, Cost, and Availability Tradeoffs in Decentralized OSNs. In WOSN’09: Proc. of the 2rd [18] F. Dabek, B. Zhao, P. Druschel, J. Kubiatowicz, and ACM SIGCOMM Workshop on Online Social I. Stoica. Towards a Common API for Structured Networks, Barcelona, Spain, August 2009. Peer-to-Peer Overlays. In IPTPS’03: Proc. of the 3rd International Workshop on Peer-to-Peer Systems, [33] R. van Rennesse, K. Birman, D. Dumitriu, and Berkeley, CA, USA, February 2003. W. Vogels. Scalable Management and Data Mining using Astrolabe. In IPTPS’02: Proc. of the 1st [19] P. T. Eugster, R. Guerraoui, A. M. Kermarrec, and International Workshop on Peer-to-Peer Systems, L. Massoulié. From Epidemics to Distributed Cambridge, MA, USA, March 2002. Computing. IEEE Computer, 37:60–67, 2004. [34] M. Zaharia and S. Keshav. Gossip-based Search [20] M. Jain and C. Dovrolis. End-to-End Available Selection in Hybrid Peer-to-Peer Networks. In Bandwidth: Measurement Methodology, Dynamics, IPTPS’05: Proc. of the 4th International Workshop and Relation with TCP Throughput. In on Peer-to-Peer Systems, Ithaca, NY, USA, February SIGCOMM’02: Proc. of the 8th ACM SIGCOMM, 2005. Pittsburgh, PA, USA, August 2002. [35] D. Zhao and M. B. Rosson. How and Why People [21] A. Java, X. Song, T. Finin, and B. Tseng. Why We Twitter: The Role that Micro-blogging Plays in Twitter: Understanding Microblogging Usage and Informal Communication at Work. In GROUP’09: Communities. In Proc. of the Joint 9th WebKDD and Proc. of the 7th International Conference on 1st SNA-KDD Workshop on Web Mining and Social Supporting Group Work, Sanibel Island, FL, USA, Network Analysis, San Jose, CA, USA, August 2007. May 2009. [22] A. M. Kermarrec, L. Massoulié, and A. J. Ganesh.
You can also read